Recurrent pregnancy loss is associated with a pro...

ARTICLE

Recurrent pregnancy loss is associated witha pro-senescent decidual response during theperi-implantation windowEmma S. Lucas1,2,5, Pavle Vrljicak1,2,5, Joanne Muter1,2, Maria M. Diniz-da-Costa1,2, Paul J. Brighton2,

Chow-Seng Kong2, Julia Lipecki3, Katherine J. Fishwick2, Joshua Odendaal2, Lauren J. Ewington 2,

Siobhan Quenby1,2, Sascha Ott1,4 & Jan J. Brosens 1,2*

During the implantation window, the endometrium becomes poised to transition to a preg-

nant state, a process driven by differentiation of stromal cells into decidual cells (DC).

Perturbations in this process, termed decidualization, leads to breakdown of the feto-

maternal interface and miscarriage, but the underlying mechanisms are poorly understood.

Here, we reconstructed the decidual pathway at single-cell level in vitro and demonstrate that

stromal cells first mount an acute stress response before emerging as DC or senescent DC

(snDC). In the absence of immune cell-mediated clearance of snDC, secondary senescence

transforms DC into progesterone-resistant cells that abundantly express extracellular matrix

remodelling factors. Additional single-cell analysis of midluteal endometrium identified DIO2

and SCARA5 as marker genes of a diverging decidual response in vivo. Finally, we report a

conspicuous link between a pro-senescent decidual response in peri-implantation endome-

trium and recurrent pregnancy loss, suggesting that pre-pregnancy screening and interven-

tion may reduce the burden of miscarriage.

https://doi.org/10.1038/s42003-020-0763-1 OPEN

1 Tommy’s National Centre for Miscarriage Research, University Hospitals Coventry & Warwickshire, Coventry CV2 2DX, UK. 2 Division of BiomedicalSciences, Clinical Sciences Research Laboratories, Warwick Medical School, University of Warwick, Coventry CV2 2DX, UK. 3 School of Life Sciences, GibbetHill Campus, University of Warwick, Coventry CV4 7AL, UK. 4Department of Computer Science, University of Warwick, Coventry CV4 7AL, UK. 5Theseauthors contributed equally: Emma S. Lucas, Pavle Vrljicak. *email: [email protected]

COMMUNICATIONS BIOLOGY | (2020) 3:37 | https://doi.org/10.1038/s42003-020-0763-1 | www.nature.com/commsbio 1

1234

5678

90():,;

http://orcid.org/0000-0003-0805-6845

http://orcid.org/0000-0003-0805-6845

http://orcid.org/0000-0003-0805-6845

http://orcid.org/0000-0003-0805-6845

http://orcid.org/0000-0003-0805-6845

http://orcid.org/0000-0003-0116-9329

http://orcid.org/0000-0003-0116-9329

http://orcid.org/0000-0003-0116-9329

http://orcid.org/0000-0003-0116-9329

http://orcid.org/0000-0003-0116-9329

mailto:[email protected]

www.nature.com/commsbio


Approximately 15% of clinical pregnancies result in mis-carriage1, most often during the first trimester. Foetalchromosomal abnormalities account for 50–60% of

sporadic miscarriages2, although the incidence is lower inrecurrent pregnancy loss (RPL)3,4, defined as two or morelosses5,6. Further, with each additional miscarriage, the frequencyof euploid loss increases whereas the likelihood of a successfulpregnancy decreases7, indicating that uterine factors drive higher-order miscarriages. Few interventions improve live-birth rates inRPL5, reflecting that in most cases the underlying mechanismsare incompletely understood.

Following the postovulatory rise in circulating progesteronelevels, the endometrium becomes transiently receptive to embryoimplantation during the midluteal phase of the cycle. Thisimplantation window heralds the start of intense tissue remo-delling8, driven by differentiation of endometrial stromal cells(EnSC) into decidual cells (DC) and accumulation of uterinenatural killer (uNK) cells9. Upon embryo implantation, DCrapidly encapsulate the conceptus10, engage in embryo bio-sensoring11, and then form a decidual matrix that controls tro-phoblast invasion12. At a molecular level, decidual transformationof EnSC encompasses genome-wide remodelling of the chromatinlandscape13, reprogramming of multiple signalling pathways14–16,and activation of decidual gene networks17,18. This multistepdifferentiation process starts with an evolutionarily conservedacute cellular stress response19, marked by a burst of reactiveoxygen species (ROS) and release of proinflammatorycytokines9,20,21. After a lag period of several days, EnSC lose theirfibroblastic appearance and emerge as secretory DC with abun-dant cytoplasm and prominent endoplasmic reticulum8. A hall-mark of DC is their resistance to multiple stress signals. Severalmechanisms underpin decidual stress resistance, includingsilencing of the c-Jun N-terminal kinase (JNK) pathway andupregulation of various stress defence proteins and ROSscavengers8,15,22. In addition, DC highly express 11β-hydroxysteroid dehydrogenase type 123, which converts inactivecortisone into cortisol, a potent anti-inflammatory glucocorticoid.Thus, compared with EnSC, DC are exquisitely adapted towithstand the hyperinflammation and stress associated with deephaemochorial placentation.

Recently, we demonstrated that decidualization also results inthe emergence of senescent decidual cells (snDC), both in vitroand in vivo9,24. In the stroma, the abundance of cells expressingp16INK4, a tumour suppressor and canonical senescence marker,peaks transiently during the midluteal phase before rising againprior to menstruation9. Cellular senescence is defined by a state ofpermanent cell-cycle arrest and prominent secretion of variousbioactive molecules, including ROS, extracellular matrix (ECM)remodelling proteins, proinflammatory cytokines, chemokinesand growth factors, referred to as senescence-associated secretoryphenotype (SASP)25–27. Different types of senescent cellsunderpin pathological and physiological processes. Chronicsenescent cells accumulate progressively in response to variousstressors and cause gradual loss of organ function during ageingmediated by the deleterious effects of the SASP on tissuehomoeostasis25,27. By contrast, acute senescent cells are linked tobiological processes that involve programmed tissue remodelling,including embryogenesis and wound healing27–29. They areinduced in response to specific signals, produce a transient SASPwith defined paracrine functions, and are promptly cleared byimmune cells25. Recently we demonstrated that snDC exhibithallmarks of acute senescent cells9. First, DC and snDC bothemerge in response to FOXO1 activation, a pivotal decidualtranscription factor downstream of the protein kinase A (PKA)and progesterone signalling pathways. Second, their associatedSASP critically amplifies the initial decidual inflammatory

response, which not only drives differentiation of EnSC but is alsolinked to induction of key receptivity genes21. Further, we haveprovided evidence that DC recruit and activate uNK cells, whichin turn may eliminate snDC through perforin- and granzyme-containing granule exocytosis9,24.

Although snDC constitute a relatively minor and variablestromal population in midluteal endometrium, they have thepotential to impact profoundly on the unfolding decidualresponse in a manner that either promotes or precludes preg-nancy progression. For example, a transient SASP associated withacute senescent cells has been shown to promote tissue plasticityby expanding resident progenitor populations9,30. By contrast,senescent cells that persist (i.e. chronic senescent cells) can inducesenescence in neighbouring cells though juxtracrine signalling(termed ‘secondary’ or ‘bystander’ senescence), leading to spa-tiotemporal propagation of the phenotype and loss of tissuefunction25,31,32.

Recently we reported loss of clonal mesenchymal stem-likecells (MSC) in midluteal endometrium of RPL patients but howthis is linked to subsequent breakdown of the decidual–placentalinterface in pregnancy is unclear33. We hypothesized that MSCdeficiency may drive a pro-senescent decidual response duringthe peri-implantation window, culminating in chronic inflam-mation in early pregnancy, proteolysis of the decidual–placentalinterface, and miscarriage. To test this conjecture, we first per-formed single-cell transcriptomic analysis on decidualizing pri-mary EnSC cultures, mapped the emergence of DC and snDC,and identified co-regulatory gene networks underpinning themultistep decidual pathway. We then extended the single-cellanalysis to peri-implantation endometrial biopsies and screenedfor stroma-specific marker genes that signal divergence of thedecidual response towards a pro-senescent decidual state. Theexpression of two marker genes, DIO2 and SCARA5, and theabundance of uNK cells were then analysed in peri-implantationendometrial biopsies from 89 RPL patients and 90 controlsubjects.

ResultsSingle-cell analysis of the decidual pathway in vitro. To identifyputative marker genes of DC and snDC, we first reconstructed thedecidual pathway in vitro using single-cell transcriptomics. Asdepicted in Fig. 1a, primary EnSC were decidualized with aprogestin (medroxyprogesterone acetate, MPA) and a cyclicadenosine monophosphate analogue (8-bromo-cAMP, cAMP)for 8 days. The differentiation signal was then withdrawn for2 days to assess progesterone/cAMP-dependency of decidualsubsets. Cells were recovered every 48 h and subjected to single-cell analysis using nanoliter droplet barcoding and high-throughput RNA-sequencing34. Approximately 800 cells weresequenced per timepoint, yielding on average 1282 genes per cell.After computational quality control (Supplementary Fig. 1), 4580cells were assigned to 7 transcriptional cell states using SharedNearest Neighbour (SNN) and t-Distributed Stochastic Neigh-bour Embedding (t-SNE) methods. Figure 1b shows cells colour-coded by day of treatment and Fig. 1c by transcriptional state(S1-7). The top 10 differentially expressed genes (DEG) betweenthe 7 cell states are presented as a heatmap (Supplementary Fig. 2& Supplementary Data 1). Apart from a discrete population ofproliferative cells (S1), the bulk of undifferentiated EnSC were inS2. By day 2 of decidualization, most cells had transitioned to S3,which differed from S2 by 898 DEG (Supplementary Data 2).This precipitous transcriptomic response is in keeping with anacute cellular stress response and coincided with a transient risein IL-6 and IL-8 secretion independent of transcription (Sup-plementary Fig. 3). Decidualizing cells then progressed in

ARTICLE COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-020-0763-1

2 COMMUNICATIONS BIOLOGY | (2020) 3:37 | https://doi.org/10.1038/s42003-020-0763-1 | www.nature.com/commsbio


−50

−25

0

25

−50 −25 0 25 50 75t-SNE1

t-SN

E2

S1S2S3S4S5/DCS6/snDCS7

−50

−25

0

25

−50 −25 0 25 50 75

t-SN

E2

D0D2D4D6D8WD

b

c

d

g

S2

S1S3

S4S5

S6

S7

Decidual cells (DC) Senescent decidual cells (snDC)

CALRVIMRSPO3TIMP3LIMS1TGM2HSP90B1PDIA6HSPA5CKAP4RPS6WNT5ACALUTIMP2ACTBMT−RNR2IGFBP3DIO2LUMABI3BPNPC2CLUSOX4GLULKIAA1199ALDH1A1IGF2IGFBP1ADAMTS5LDLRLAYNPCDH20CXCL14ING1ADAMTS1FADS1CALM2SPARCL1GLRXSCARA5CTSLFERMT2MCFD2FTLPGRMC2IL1RL1EDNRBCOL14A1ARPC2NDUFB2

Cell Type

−3

−2

−1

0

1

2

3

Gen

e ex

pres

sion

leve

lDecidualization Senescence

Pre-branch

P = 4.29 × 10-52 P = 5.20 × 10-09 P = 3.07 × 10-29

P = 5.85 × 10-51

S5 S6

Exp

ress

ion

leve

l per

cel

l

0123

0123

0

2

4

6

01234

S5 S6

GLRX

ABI3BP

HSD11B1CRYAB

01234

KIAA1199

S5 S6

P = 9.14 × 10-49

Timepoint

Cell state

f

ADAMTS5

01234

P = 8.06 × 10-29

C+M withdrawn

8-bromo-cAMP and MPA (C+M)Single cell analysis

Media change

D0

D2

D4D6

D8WD

0 48 96 144 192 240

Time (hours)a Endometrial

biopsy

Stromalfraction

Passage 1

Passage 2

Com

pone

nt 2

Component 1

1

−3

−2

−1

0

1

2

−5 0 5

S2 S3 S4

DC snDC

0102030405060708090

100

D0 D2 D4 D6 D8 WD

DCsnDCS7

Pro

porti

on o

f cel

ls (%

)

e

Fig. 1 Single-cell reconstruction of the decidual pathway in culture. a Schematic representation of in vitro time-course experiment. b t-SNE projection of4580 EnSC, colour-coded according to days of decidualization (D0-8) and upon withdrawal (WD) of the differentiation signal for 48 h. D0 representsundifferentiated EnSC. c The same t-SNE plot now colour-coded according to transcriptional state (S1–7). d Violin plots showing log-transformed,normalized expression levels for indicated genes in decidual cells (DC, state S5) and senescent decidual cells (snDC, state S6). e Relative proportion oftotal cells assigned to states S5 (DC), S6 (snDC) and S7 at each experimental timepoint. f Decidualizing EnSC were placed in pseudotime to reconstructthe trajectory of differentiation, revealing a continuous trajectory towards senescence with a single branch point marking the divergence of DC. g Heatmapshowing gene dynamics during cell state transition at the branch point shown in panel (f). Columns are points in pseudotime while rows represent the 50most dynamic genes at the branch point. The beginning of pseudotime is in the middle of the heatmap and the trajectory towards DC and snDC areindicated by the arrows. Hierarchical clustering visualizes modules of genes with similar lineage-dependent expression patterns.

COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-020-0763-1 ARTICLE




synchrony to S4 by day 4 after which they segregated into twotranscriptionally distinct subsets, S5 and S6 (Fig. 1c). Analysis ofindependent EnSC cultures confirmed that decidualizationpolarizes cells into two subsets (Supplementary Fig. 4). Knowndecidual stress defence genes (e.g. CRYAB, HSD11B1 andGLRX)35–37 were enriched in S5 (designated DC) whereas genesinvolved in oxidative stress signalling and cellular senescence,including KIAA1199, CLU and ABI3BP38–41, prevailed in S6(designated snDC) (Fig. 1d and Supplementary Fig. 5). Notably,on day 6, the proportion of DC and snDC was 78% and 13%,respectively (Fig. 1e and Supplementary Table 1). By day 8, snDCwere almost as abundant as DC (42% and 45%, respectively),suggesting rampant secondary senescence. Withdrawal of thedecidualization signal on day 8 elicited a further transcriptionalchange (S7), which was accounted for by partial de-differentiationof >95% of DC (Supplementary Data 2 and SupplementaryTable 1). By contrast, snDC were much more refractory towithdrawal, suggesting loss of progesterone/cAMP-dependency.

To explore the ‘switching’ of DC into snDC further, we ordereddecidualizing cells (D2-8) in pseudotime and colour-coded themby transcriptional state. As shown in Fig. 1f, this analysis revealedthat decidualizing EnSC progress along a continuous trajectorytowards senescence. This trajectory was interrupted by a singlebranch point, marking the divergence of DC and a subset of DCalready transitioning towards a senescent phenotype. The top 50genes underpinning this branch point are depicted in a modifiedheatmap and clustered hierarchically to visualize modules ofgenes with similar expression patterns (Fig. 1g). DIO2, codingiodothyronine deiodinase 2, was identified as a major branch genein the senescent pathway. This enzyme catalyses the conversionof prohormone thyroxine (T4) into bioactive triiodothyronine(T3)42, suggesting increased energy metabolism in snDC. Multi-ple genes coding for senescence-associated ECM remodellingfactors were co-regulated with DIO2, including LUM (lumican)43,CLU (clusterin)41, ADAMTS5 (ADAM metallopeptidase withthrombospondin type 1 motif 5)44, KIAA1199 (also known as cellmigration inducing hyaluronidase 1, CEMIP)39 and ABI3BP (ABIfamily member 3 binding protein)38. Notably, IGFBP1, a widelyused decidual marker gene8, was also part of this module. FTLand SCARA5, along with known decidual genes such as GLRXand IL1RL1, were part of a prominent branching module in thenon-senescent decidual pathway (Fig. 1g). FTL encodes ferritinlight chain (L-ferritin) and SCARA5 (scavenger receptor class Amember 5) the L-ferritin receptor, suggesting increased ironstorage and detoxification capacity in DC.

Taken together, the single-cell analysis confirmed thatdecidualization is a multistep process that starts with an acutestress response, which in turn synchronizes transition of cellsthrough intermediate transcriptional states before emergingmainly as DC and some snDC. We also demonstrated that snDCrapidly perpetuate the senescent phenotype across the culture,resulting in chronic senescence and increased expression of ECMconstituents and proteases and other SASP components.

Co-regulated decidual gene networks. We used k-means clusteranalysis to identify networks of co-regulated genes across thedecidual pathway (Supplementary Data 3). Analysis of 1748 DEGyielded 7 networks of uniquely co-regulated genes. Figure 2depicts individual networks annotated for selected transcriptionfactor (TF) genes with core roles in decidualization. Network A1genes are rapidly downregulated within the first 48 h of thedecidual process after which expression remains largely stable.The most notable TF to be ‘reset’ in this manner upon decid-ualization is the progesterone receptor (PGR). This observation isin keeping with a previous study purporting that overexpressed

PGR blocks the formation of multimeric transcriptional com-plexes upon decidualization by squelching key co-regulators45.Network A2 genes, including HOXA10, are progressively down-regulated upon decidualization. Networks B1 and B2 comprise ofbiphasic genes that peak prior to the emergence of DC and snDC.They include several genes coding for pivotal decidual TFs, suchas STAT3, MEIS1, KLF9, WT1 and FOXO18,19. Genes in networkC1 are gradually induced upon decidualization and then peak inDC. Prominent TFs in this network are heart and neural crestderivatives expressed 2 (HAND2) and FOS like 2 (FOSL2), apotent PGR co-regulator46. By contrast, two distinct networksunderpinned the emergence of snDC. Network C2 genes risegradually during the initial decidual phase but expression is thenrapidly accelerated, especially in snDC. Interestingly, this networkis enriched in gene ontology terms ‘secreted’ (Benjamini adjustedP= 3.7 × 10−5, modified Fisher Exact test) and ‘type I interferonsignalling pathway’ (Benjamini adjusted P= 1.4 × 10−6, modifiedFisher Exact test) (Supplementary Data 4), both canonical fea-tures of cellular senescence24. A prominent TF in this senescence-associated network is signal transducer and activator of tran-scription 1 (STAT1), which is not only activated by interferonsignalling but is also a potent inhibitor of PGR signalling inendometrial cells47. A second biphasic network, designated net-work D, is also associated with snDC. Genes in this network areinitially repressed upon decidualization and then re-expressedpredominantly in snDC. Intriguingly, this network is enriched ingenes known to be repressed by PGR, including the TF genesSOX4 and FOXP114. Taken together, the data show that complexnetworks of co-regulated genes underpin progression of cellsalong the decidual pathway and suggest that altered expressionlevels of PGR co-regulators, such as FOSL2 and STAT1, mayaccount for progesterone-dependency of DC and progesterone-resistance of snDC.

Coordinated activation of immune surveillance genes. Wemined the networks further for genes encoding factors implicatedin immune surveillance of senescent cells. CXCL14 and IL-15drive uNK-cell chemotaxis and proliferation and activation,respectively9,48. TIMP-3 (TIMP metallopeptidase inhibitor 3)plays a critical role in immune recognition of stressed or senes-cent cells by inhibiting proteolytic cleavage of surface stressligands needed for NK cell recognition49. Intriguingly, CXCL14,IL-15 and TIMP-3 belong to the same biphasic gene network(B2), characterized by peak expression immediately prior to theemergence of DC and snDC (Fig. 3a). Subsequently, expression ofthese genes drops markedly in snDC but much less so in DC. Toexplore this concept of ‘programmed’ immune surveillance fur-ther, we monitored the secreted levels of CXCL14, IL-15 andTIMP-3 every 48 h over an 8-day time-course in four indepen-dent decidualizing cultures. As shown in Fig. 3b, secreted levels ofall 3 factors rise quickly during the initial decidual phase. Whilethe levels of CXCL14 and TIMP-3 then appear to plateau, IL-15continues to accumulate in the supernatant.

Clusterin (CLU) and the soluble IL-33 decoy receptor sST2(encoded by IL1RL1) are putative secreted markers of snDC andDC, respectively (Fig. 1g). As shown in Fig. 3c, secreted levels ofboth markers increase markedly upon decidualization, althoughthe rise in sST2 levels generally preceded CLU secretion. Wespeculated that uNK cells would selectively attenuate CLUsecretion by targeting snDC. To test this hypothesis, uNK cellswere isolated from the supernatants of freshly cultured EnSCusing magnetic-activated cell sorting (MACS) and the purity andviability of cells confirmed by flow cytometry (SupplementaryFig. 6). The experimental design is depicted in Fig. 3d. Tomonitor uNK-cell killing of senescent cells, we measured




senescence-associated β-galactosidase (SAβG) activity, a widelyused senescence marker26, in undifferentiated and decidualizingcultures with or without uNK cells. As reported previously9, theincrease in SAβG activity upon decidualization was entirelyabrogated by co-cultured uNK cells (Fig. 3e). Further, uNK cellsreduced the secreted levels of CLU (Fig. 3f), whereas the impacton sST2 levels was minimal (Fig. 3g). To confirm that uNK cellstarget snDC, six independent primary EnSC cultures weredecidualized for 8 days in the presence or absence of dasatinib,a potent senolytic drug9. As shown in Supplementary Fig. 7,dasatinib also markedly reduced SAβG activity and selectivelyinhibited CLU whereas the effect on sST2 secretion was minimal,thus recapitulating the actions of uNK cells. Taken together, thedata show that the expression and secretion of decidual factorsinvolved in uNK-cell recruitment and activation is hardwired tocoincide with the emergence of snDC. If recapitulated in vivo,these observations further suggest an important role for uNK cellsin limiting the detrimental impact of snDC in early pregnancy.

Single-cell analysis of peri-implantation endometrium. Decid-ualization involves activation of numerous genes that are eitherconstitutively expressed or induced in glandular epithelial cellsfollowing ovulation8. To identify stroma-specific marker genes of adivergent peri-implantation decidual response, we subjectedfreshly isolated endometrial cells to scRNA-seq. Biopsies weretimed relative to the pre-ovulatory luteinizing hormone (LH) surgeto coincide with the midluteal implantation window (LH+ 8;n= 3) or the start of the late-luteal phase (LH+ 10; n= 3).Following quality control (Supplementary Fig. 8), t-SNE analysisassigned 2847 cells to 5 clusters colour-coded to indicate the dayof biopsy (Fig. 4a). Additional dimensionality reduction analysiswas performed on immune cells (Fig. 4b). Clusters were desig-nated based on canonical marker genes as endothelial cells (EC;n= 141), epithelial cells (EpC; n= 395), immune cells (IC; n=352) and EnSC (n= 1943) (Fig. 4c and Supplementary Data 5). Inaddition, a discrete but as yet uncharacterized cluster of pro-liferating cells (PC; n= 16) was identified (Fig. 4a). EpC segre-gated in 4 clusters with the most abundant population (EpC1)expressing prototypic receptivity genes (e.g. GPX3, PAEP andDPP4) (Fig. 4c)50. EpC2 are ciliated epithelial cells found inter-spersed throughout endometrial glands (Supplementary Fig. 9).EpC3 were derived predominantly, but not exclusively, from asingle biopsy whereas EpC4 represented an ambiguous popula-tion expressing both epithelial and stromal marker genes(Fig. 4c). uNK cells, representing 89% of all IC, clustered intothree subpopulations (NK1-3; Fig. 4b). Notably, different NKpopulations have also recently been described in pregnant

decidua51. Based on CIBERSORT analysis, the remainingimmune populations were identified as naive B-cells (IC1),monocytes (IC2), and macrophage/dendritic cells (IC3) (Fig. 4dand Supplementary Data 6). The abundance of various cell typesand subsets in each biopsy is tabulated in Supplementary Table 1.

Marker genes of diverging decidual states. Next, we focused onthe EnSC, which clustered prominently by day of biopsy in thet-SNE plot (Fig. 4a). Progression from LH+ 8 to LH+ 10 wasassociated with altered expression of 518 genes in EnSC (Sup-plementary Data 7), 49% of which are also part of the 7 co-regulated gene networks in vitro (P < 10−143, hypergeometrictest). Although our single-cell analysis was limited to six biopsiesobtained on only two time-points, we surmised that the rela-tionship between genes that drive the divergence of DC and snDCin vitro would, at least partly, be conserved in stromal cellsin vivo. To test this hypothesis, we selected the top 50 genes of thebranch point in the decidual time-course and determined thePearson correlations for each gene pair in EnSC in vitro andin vivo on LH+ 8 and LH+ 10, respectively. We then calculatedthe sum of coefficients, ranging from +2 to −2. Congruency wasdefined as the sum of correlation coefficients of >1 or <−1 forpositively and negatively co-regulated genes, respectively (Sup-plementary Fig. 10). Using this criterion, 27% of gene pairs werecongruent on both LH+ 8 and LH+ 10, which is significantlymore than expected by chance alone (P < 10−30 and P < 10−58,respectively; hypergeometric test).

We reasoned that informative marker genes of a divergentdecidual response in vivo should be highly enriched in stromalcells, not regulated in glandular epithelium, and have a temporalprofile across the luteal phase commensurate with the expectedswitch of DC to snDC prior to menstruation. Out of 50 branchgenes, 5 genes (TIMP-3, IGF2, DIO2, SCARA5 and ABI3BP) werehighly enriched in EnSC compared with EpC, EC or IC. Whencross-referenced against two publicly available datasets [GeneExpression Omnibus (GEO) ID: GSE84169 and GEO ID:GDS2052), only SCARA5 and DIO2 met all criteria. Notably,SCARA5 belongs to the C1 network of genes whose expressionpeaks in DC whereas DIO2, a gene repressed by the activated PGRin decidualizing EnSC14, belongs to the D network of senescence-associated genes (Fig. 5a). We examined the relative expression ofboth genes in different endometrial cell types (Fig. 5b) as wellas their temporal expression across the cycle (Fig. 5c). Next, wemeasured SCARA5 and DIO2 transcript levels by RT-qPCR in250 samples obtained across the implantation window (LH+6–11) to generate percentile graphs based on the statisticaldistribution in gene expression for each day (Fig. 5d). To

PGR

0

25

50

75

100

S2 S3 S4 DC snDC

Nor

mal

ised

exp

ress

ion A1

HOXA10

A2

STAT3MEIS1KLF9

B1

FOXO1WT1

B2

FOSL2HAND2

0

25

50

75

100C1

STAT1C2

FOXP1SOX4

D

S2 S3 S4 DC snDC S2 S3 S4 DC snDC

S2 S3 S4 DC snDCS2 S3 S4 DC snDC S2 S3 S4 DC snDC S2 S3 S4 DC snDC

Cell State Cell State Cell State

Cell State

Nor

mal

ised

exp

ress

ion

0

25

50

75

100

Nor

mal

ised

exp

ress

ion

0

25

50

75

100

Nor

mal

ised

exp

ress

ion

0

25

50

75

100

Nor

mal

ised

exp

ress

ion

0

25

50

75

100

Nor

mal

ised

exp

ress

ion

0

25

50

75

100

Nor

mal

ised

exp

ress

ion

Fig. 2 K-means cluster analysis identifies co-regulated decidual gene networks. Analysis of 1748 DEG across the decidual pathway (D0–D8) yieldedseven networks of uniquely co-regulated genes. Networks are annotated for selected transcription factor genes with core roles in decidualization.





a

b

c

IL15

TIMP3CXCL14

0

1

2

3

CX

CL1

4 (n

g/m

l)

C+M (days)0 2 4 6 8

C+M (days)0 2 4 6 8

0

50

100

150

IL-1

5 (p

g/m

l)

S2 S3 S4 DC snDC0

25

50

75

100

Nor

mal

ised

exp

ress

ion

d

020406080

TIM

P-3

(ng/

ml)

C+M (days)0 2 4 6 8

B2

100

0

2

4

6

0 2 4 6 8

sST2

(ng/

ml)

0

200

400

600

800

0 2 4 6 8

CLU

(ng/

ml)

C+M (days)C+M (days)

e

0

20000

40000

60000

80000

100000

SA

βG (F

IU)

C+MuNK cells +

a

b

aa

+++

-- -

-

D2 D4 D6

Treated with C+M

Undifferentiated

uNK Cells

SAβG,CLU, sST2

SAβGD2

D8

0

200

400

600

800

1000

0

1

2

3

4

CLU

(ng/

ml)

sST2

(ng/

ml)

uNK cells +- uNK cells +-

f g

Fig. 3 Coordinated expression of decidual immune surveillance genes. a Decidual gene network B2 annotated to highlight genes implicated in uNK-cellactivation and immune surveillance of senescence cells. b Primary EnSC cultures were decidualized with 8-bromo-cAMP and MPA (C+M) for theindicated days. ELISAs were performed on spent medium collected at 48 h intervals to examine secreted levels of CXCL14, IL-15 and TIMP-3. Grey dottedlines indicate secreted levels in individual cultures (n= 4). Black solid lines indicate the median level of secretion. c Four independent primary EnSCcultures were decidualized with 8-bromo-cAMP and MPA (C+M) for the indicated days. ELISAs were performed on spent medium collected at 48 hintervals to examine secreted levels of clusterin (CLU) and sST2. Cultures established from the same biopsy are colour matched between plots (dottedlines). Black solid lines indicate the median level of secretion. d Schematic representation of uNK-cell co-culture experiments. A total of 5000 primary uNKcells were co-cultured for 48 h from the indicated timepoint with 50,000 decidualized cells seeded in 96-well plates. Four independent EnSC cultures wereused. e SAβG activity in undifferentiated and decidualized (C+M) cells co-cultured with or without uNK cells. Four independent EnSC cultures were used.f Secretion of CLU in decidualized (C+M) cells co-cultured with uNK cells relative to decidualized cells cultured without uNK cells. g Secreted sST2 levelsin the same culture. Four independent EnSC cultures were used; lines connect paired samples.




determine if SCARA5 and DIO2 transcripts are co-expressed ormark different decidual cells, we performed multiplexed single-molecule in situ hybridization (smISH) on endometrial biopsiesobtained on the same cycle day but deemed SCARA5HIGH/DIO2LOW, SCARA5LOW/DIO2HIGH or SCARA5AVERAGE/DIO2AVERAGE based on the corresponding percentile graphs(Fig. 5e). Most EnSC were SCARA5+ but DIO2− in SCARA5HIGH/DIO2LOW biopsies whereas the opposite pattern was observed inSCARA5LOW/DIO2HIGH samples. However, SCARA5AVERAGE/DIO2AVERAGE samples consisted of mixture of SCARA5+ orDIO2+ cells as well as intermediate cells expressing bothtranscripts. Finally, we used a computational approach to isolateEnSC relatively enriched in SCARA5 transcripts but reducedDIO2 expression and vice versa. As shown in Supplementary

Fig. 11, SCARA5enriched/DIO2reduced cells are characterized by theexpression of multiple decidual TF and stress-resistance geneswhereas a stress gene signature was prominent in SCARA5reduced/DIO2enriched cells (Supplementary Data 8).

RPL is associated with an aberrant decidual response. Toexamine if an aberrant decidual response in cycling endometriumis linked to increased risk of pregnancy loss, we analysed LH-timed endometrial biopsies from 90 control subjects and 89 RPLpatients. Demographic and clinical characteristics are presentedin Table 1. Each biopsy was divided and processed for RT-qPCRanalysis and immunohistochemistry. SCARA5 and DIO2 tran-script levels were measured and normalized for the day of the

EC

EnSC

EpC1

EpC2

EpC3*

EpC4

PC

IC

-40

-20

0

20

−20 0 20 40

a

c

t-SN

E2

IC1

IC2

IC3

NK

1

NK

2

NK

3

CXCR4BTG1CCL5TIGITPRF1CD3DEPAS1GNLYANLNMKI67TOP2AIL2RBGZMANCAM1CD96ITGA1C1QBCCL20CXCL3APOC1GPNMBIL8CXCL2CD14HLA−DRAHLA−DPA1HLA−DRB1CD4CD74AIF1CD83LY9CD79AMS4A1CD19

−2 −1 0 1 2Row Z−Score

EC

EnS

C IC

EpC

1

EpC

2

EpC

3

EpC

4

−2 −1 0 1 2Row Z−Score

Color Key

d

C1SMMECOL3A1SPATA22CMC2PLAURANXA2CD55PDK4FTH1S100A10CAPSAGR3FHAD1DNAAF1RSPH1PAEPCD36DPP4CXCL14GPX3ALCAMRASD1MYO6CD4CD19PTPRCMEG3ZEB1DCNMCAMTIE1A2MANGPT2

t-SNE1

LH+10LH+8

−10

−5

0

5

10

15

−10 0 10 20

IC1

IC2

IC3NK1NK2

NK3

t-SNE1

t-SN

E2

Immune cells

Color Key

−40

b

Fig. 4 Identification of endometrial cell types and subsets during the implantation window. a t-SNE plot of 2847 cells isolated from six LH-timedbiopsies, coloured to indicate reported cycle day (LH+ 8 or LH+ 10), captures all major endometrial cell types, including epithelial cells (EpC), immunecells (IC), endothelial cells (EC), stromal cells (EnSC) and a discrete but transcriptionally distinct proliferative (PC) stromal subpopulation. EpC segregatedin four subpopulations (EpC1-4). * indicates EpC3 contributed predominantly by a single sample. b additional dimensionality reduction separated immunecells into three uNK-cell subsets (NK1-3), naive B-cells (IC1), monocytes (IC2) and macrophage/dendritic cells (IC3). c Heatmap showing relativeexpression (z-score) of markers defining cell types and EpC subpopulations. MYO6, RASD1 and ALCAM are included as pan-epithelial genes. d Heatmapshowing relative expression of markers defining endometrial IC populations during the implantation window, including three uNK-cell subsets.





cycle based on their respective percentile graphs (Fig. 5d).Immunohistochemistry was used to quantify uNK cells using astandardized and validated approach that involves measuring thenumber of CD56+ uNK cells per 100 cells in proximity of theluminal epithelium52. The uNK-cell data were then normalizedfor the day of the biopsy in the cycle as reported previously9. As

shown in Fig. 6a, lower SCARA5 but higher DIO2 percentiles inRPL samples indicated a diverging pro-senescent decidualresponse. Further, uNK cells were significantly less abundant inRPL compared with control subjects (P < 0.0001, Welch two-sided t-test). We reasoned that SCARA5+ DC and uNK cells drivesuccessful transformation of the stroma into the decidua of

● ● ●●● ●●● ● ●●

●●

●

●

●

●● ● ●●● ●●

●

● ●● ●● ● ●●● ● ●

●

●● ● ●● ●● ●● ●● ●

●●

●

●

●● ● ● ●

●

●

●

● ●●

●

● ● ●

●

●

●

● ●● ●● ●●● ● ●●●● ●● ● ●●● ● ● ● ●● ●●● ●● ● ●

●

●

●

●

●● ● ● ●● ●● ●● ●● ● ●● ● ●●● ●● ●● ●

●

●

●●● ●

●

●●● ●● ● ● ●●● ●●● ● ●●● ●● ●

●

●

●

● ●● ●●●

●

●

● ●●

●

●

●

●

●

●● ●● ●● ●●

●

●● ●●● ●●● ●●

●

●

●

●

●

● ● ●●● ●● ●

●

●

●

●

●

●●

●

● ● ●●

● ●●

●●

●

● ●●●● ●● ●● ●● ●

●

● ●●●●

●

● ● ●●● ● ●

●

●

●

● ●● ●●● ●●● ●●

● ●

● ●

●

● ●● ●● ●●

●

● ●

●● ● ●●● ●● ●●● ● ●●●● ●● ●

●

●●●

●

● ●

●

●

●

●

●●●

●●

●●● ●

●

●

● ●

●

●

● ●

●

● ●●●● ●●

●

●

●● ●● ●● ●● ●●●●

●● ●

● ●

●

● ●●● ●● ● ●● ●●●

●●

● ●

●

●

●

● ●● ●● ●●● ●

●

●● ●● ●● ●● ●●

●

● ●●

●●

●●● ●

●●

●

●

●

●

●● ●● ●●●

●

● ●

●

● ●●●

●

●

●

●

●

●

●

● ●●●● ●

●

● ●●

●

●● ● ●●

●

●

●

●

●

● ●● ● ● ●

●

●●

●

● ●● ●

● ●

● ●

●

●

●●

●

●

●●●

●

●●● ● ●● ●●● ●● ● ●●

●

●●● ●

●

● ●●

●

●● ●●

●

●● ● ●●● ●●

●

● ● ●●

●

●

●

●● ●● ●●

●

●

●

●●● ●

●

●●●

●

●● ●

●

● ●● ●●

●

●●●

●

●

● ●●

●

● ●●

●

●

●● ●●●

●

●

●

●

●

●

●●

●

●

● ●

●

● ● ●

●

●

●●

●● ●●

●

●●●

●

●

●

●

●●● ●

●●

●● ●

●

●● ●

●

●●●

●

●● ●●

●

●

●

●●●●

●

●●

●

● ●

●

●● ●

●

● ●● ●● ● ●●● ●● ●●● ●

●

● ●● ●

●

●● ●●

●

●● ● ●●● ●● ●●●● ●● ●● ●●●●

●

●

● ● ●● ● ●● ● ●●●●

●

●

● ● ●● ●●●

●

●●● ● ● ●●●

●

●

●

●●

●●

●

●

●

●●

●

●●● ●

●

● ●

●

●

●●

●

● ●

●

●

●

●●

●

●

● ●

●

●● ●● ● ●●●

●

●

●

●

●

●

●

●● ●●

●

●

●

●

●

●

●●●● ●● ● ●●●●●

●

● ●

●

●● ● ●●● ●●●

●

● ●●

●

●

●● ●● ●

●●

●●

●

●●

● ●

●●

●

●

●●

●

● ● ●

●

● ●●

●

●● ●

●

●● ●

●●

● ●● ●

●

●

●

●●

●

●● ●●

●

●●

●

●● ●●● ●

● ●

●

●

●

●

●● ● ●

●

● ●●● ●●●● ●

●

● ●● ●● ●

●

● ●●●●

●

●●●●

●

●● ●

●

●● ●

●

●●

●

● ●●● ●● ●●●●●

●

● ● ●●● ●● ● ● ●

●

●●

●

●

●

●

●

●●

●

●

●

●● ●

●●●

● ●● ●●

●

● ●●

●

● ●●●

●

● ● ● ● ●●●

●

●

●

● ●● ●●●●

●

●● ●●●

●

●●

●

●

●

●

●

●●●

●

●

●● ●●●●

●

●●●● ●

●

●

●

●●● ●●

●

●

●

●● ●●

●

● ●● ●●●● ● ●

●

●

●

● ● ●●

●

●

●● ●●

●

●

●

●●

●

●●

●

●

●●● ●●●●● ●

●

●●

●

●● ●●●●●● ●●● ● ●●

●

●● ●● ●● ●●

●

●

●

●● ●●●

●●

●●● ●● ●● ●● ●● ●●

●

●● ● ●● ●● ● ●● ●● ●● ● ●●● ●

●

● ● ●● ●● ● ●●● ●

●

●●●● ● ●●●● ●●●● ●

●

●●●● ●●● ●

●

● ●●●●

●

●●

●

●●● ●● ●● ●

●

●● ●●● ●● ● ●●●● ●●● ●● ● ● ●● ●●●● ●● ●●● ● ●● ●

●

●● ● ●● ● ●●● ●●

●

● ●●● ●● ●●

●

● ●

●

●

●

● ● ●●

●

●

●

●

●

●● ●●●

●

●

● ●●● ● ●● ●●● ●●

●

●

●

●

● ●

●

● ●●● ● ●

●

● ●

●

●●

●● ●●

●

● ●●

●

● ●●●

●

●

●

●●● ● ●

●

● ●

●

●

●

●

● ●

●● ●

●

●

●

●● ●● ●● ●

●

● ●●

●

● ●●●

●●

●●

●

● ●●● ● ● ●

●

● ●●● ●●

●

● ●● ●● ●

●

●● ●● ●

●

● ●● ●● ● ●

●

●● ●

●

●● ●●● ●●● ●● ● ●● ●● ●● ● ● ●

●

●

●

●●● ●

●

● ● ●

●

●●● ● ●●●● ●●● ●●●●● ● ●● ●● ● ●

●

●●● ●●●

●

●● ●● ●●

●

●

●●● ●

●

● ●● ●●●●

●

● ●●

●

● ●

●

● ●● ●●● ●●

●

●

●

●

●

●

● ●

●

● ●

●

● ● ●

●

●● ●●● ●

●

●

●●●

●●●

●

●●

●

●

●●

●

●

●●

●

●

●●

●

●

●

●●●

●

●

●

● ●●● ●●

●

●

●

●

●●●● ●●

●

●

●●

●

●

●

● ●●● ●● ●

●

●

●●

●●

●

●●● ● ● ●

●●

●●

●

● ●● ●

●

●

●

●

●

● ●● ● ●● ● ●

●

● ●

●

●

●

●●

●

● ●●● ●●● ●●

●

●●●

●

●●

●

● ●● ● ●●● ●

●

●● ● ●● ●● ●●● ●● ●●● ●● ●●● ●●●● ● ●

●

●●● ●●● ●●● ●

●

●●● ● ●●

●

● ●● ●● ● ●

●

●

●● ● ● ●●

●

●

●●

●

● ●●

●

● ●● ●

●

●●● ●

●●

● ●● ●● ●● ●

●

●●● ● ●● ●●● ●● ●

●

●● ●●● ●●● ●●●

●

● ● ●●

●

●●

●

● ●●●● ●

●

●●●

●

●

●●●

●

●●

●

●

●

● ●

●

●●● ●●

●

●● ●● ●●●

●●●

●

●

● ●●

●

● ●

●

●● ●

●

● ● ●● ● ●●● ●● ●● ●●

●

●● ●●● ●

●

● ●●

●

●

●● ●● ●● ●● ●

●

●

● ●● ● ●●● ● ●●

●

● ● ●●

●

●●● ●● ●●● ● ●● ●●● ●●● ● ●●●●

● ●

● ●● ●● ●●

● ●●

● ● ● ●●

●

● ●●● ●●● ●● ●●● ●●● ●

●

● ● ● ●●

●

●● ●

●

●

●

●●●● ●●

●

●●

●

● ●● ●● ●●●

●

● ●● ● ●●● ●●● ● ●● ● ●●● ●●

●

● ● ● ●●●

●

●

●

●

●

●

●●

●

●

●

●

●

●

●

●

● ●●

● ●

●●

●

●

●

●

●

●●

●

● ●●● ●

●

●● ●● ●

●

● ●

●

●

●

●

●

●

●●● ●● ●●

●●

●●

●● ●

●●●

●

●●● ●

●

●●

●

●

●

● ●● ●● ● ●●

●●

● ●● ●

●

●● ●●

●

●● ●●● ●●●

●

●

●

●

●

● ● ● ●

●

● ●

●

●●● ●●

●

● ●

●

● ●

●

●

●

● ● ●●●

●

●

●

● ● ●●● ● ●●

●●

●

●●●

●●

●

● ●

●

● ● ●● ●●

●

●

●

●

●●●● ● ● ●

●

●● ●

●

●●● ●

●

● ● ●

●

● ●● ●● ●●● ●● ●●

●

●● ●●

●

●

●

●●● ●● ●

●

● ●● ●●● ●● ● ●● ● ●●

●

●● ● ●● ●●● ●●

●

●●● ●● ●●● ●●

●

●

●

●

●

●

●

●

● ● ●

●

● ●●● ●● ●●

●

●●● ● ●●● ●●● ●●● ●● ●●●● ●

●●●

● ●●●● ●●●●

●

●

●

●

●●●

●

●●●

●●

●●● ● ● ●

●

●● ● ●● ●

●

● ●

●

● ●

●

●● ●● ●● ●

●

●

● ●● ●●

●

●

●

● ● ●● ● ●

●

●● ●●● ● ●● ● ● ●●

●

● ●

●

●● ● ●● ●●●● ●● ● ●● ●●●●

●

● ●● ●● ●

●

●● ●● ●● ●●● ●

●

●●● ● ●●●

●

● ●●●● ● ●●● ●● ●● ●● ●●●●

●

● ● ●● ● ● ●● ● ●

●

●●● ●● ●● ●●● ● ●●●

●

●

●

●● ●●● ●●●● ●● ● ● ● ●● ●●●● ●●● ●● ● ●●● ● ●● ●● ●●● ●● ● ●● ●● ●●●● ●

●

●●● ●●●● ● ● ●● ●● ●● ●● ●●● ● ●● ●● ●● ●● ● ●● ● ● ●●●●

●

● ●● ●●

●

● ●● ●●● ●●● ●

●

● ●●●● ●●● ●●● ●● ●●● ●● ●● ● ●● ●

●

● ●● ●● ● ●●● ● ●●● ●●● ● ●● ●● ●●●●● ●

●

●● ●●●● ●

●

●● ●●● ● ●●● ●● ● ●●●● ●●● ● ●●●● ●● ●● ●● ●● ●●● ●●● ● ● ● ●● ●● ● ●●●● ● ●●● ● ●●● ●●● ● ●● ●● ● ●● ●● ●●● ●● ● ●●●● ●●●●●● ●● ●● ●● ●● ● ●● ● ●●●●● ● ●

●

●● ●●● ●●●● ●● ●● ● ●● ●●● ● ●●● ● ●● ● ●●● ●●● ●●

●

●●● ●●● ●● ●● ●● ● ●●●● ●● ●● ●● ●● ● ● ●● ●● ●● ●● ●

●

●

●●● ● ●● ●● ●●● ●●● ●● ● ●● ● ●●● ●● ●● ●●● ● ●● ●● ●

●

●●● ●●● ●●● ● ● ● ●●● ●● ●● ●

●

●● ●● ●● ●● ●● ●●● ● ● ●●● ● ●●●●● ●● ●● ●●● ●● ● ●●●●●● ● ●● ●●● ● ● ●●●●● ● ● ●● ●● ●● ●● ●●● ●● ●● ●● ●● ●●● ● ●● ● ●●● ●● ●● ●● ●● ●●● ●● ●●● ●●●● ● ●●

●

● ●

●

●●

●●● ●● ● ●● ●

●

●● ●●● ●● ●●●● ● ●●● ●●● ●● ●● ●● ●● ●● ●● ● ●● ●●●● ●

●

●● ●●● ●●

●

●●● ●● ●● ● ●● ● ●●● ● ●●● ● ●● ●● ●●● ●● ●●● ● ●●● ● ● ●●● ● ●● ●

●

● ●● ●●● ●●●●●● ●● ●● ●● ●●● ● ● ●● ●●● ●● ●● ●●● ●●● ●●● ● ●●● ●● ●● ●● ●

●

● ●● ● ●● ●● ●● ●●● ●●● ●●●● ●● ● ●●● ●● ●●●● ● ●●●●

●

●

●

● ●● ●●●● ●●● ●●●●● ● ● ●●● ●●● ●●● ●●●●● ●●● ●●● ●● ●●● ● ●●●● ●● ● ● ● ●● ● ●●● ●●●● ●0

1

2

3

4

EC EnSC EpC IC

DIO2P = 2.22 x 10-27

0

1

2

3

SCARA5P = 6.12 x 10-33

b d

10

116 7 8 9 10 11

LH+dC

t

9

10

11

12

13

dCt

> 70%30−70% < 30%

SCARA5

0

100

200

300

400

500

Rel

ativ

e ex

pres

sion

(A.U

.)

0

1000

2000

3000

4000

5000 DIO2

SCARA5

Pro

lifer

a tiv

e Early MidSecretory

Exp

ress

ion

leve

l per

cel

l

7

8

9

SCARA5HIGH / DIO2LOW

e

SCARA5AVERAGE / DIO2AVERAGE SCARA5LOW / DIO2HIGH

Late

a

SCARA5

0

25

50

75

100

Nor

mal

ised

exp

ress

ion C1

DIO2

0

25

50

75

100 D

Nor

mal

ised

exp

ress

ion

S2 S3 S4 DC snDCCell state

S2 S3 S4 DC snDCCell state

c

DIO2

Fig. 5 SCARA5 and DIO2 as marker genes for divergent decidual states. a SCARA5 and DIO2 belong to two distinct decidual gene networks with peakexpression in DC and snDC, respectively. b, c Spatial and temporal expression of SCARA5 and DIO2 in cycling human endometrium: b violin plots showingexpression of SCARA5 and DIO2 in vivo in EnSC, EC (endothelial cells), EpC (epithelial cells) and IC (immune cells) (Wilcoxon rank sum test withBonferroni correction); c expression of SCARA5 and DIO2 in proliferative and early-, mid-, late-luteal phase endometrium. Each bar represents an individualbiopsy. The data were retrieved from microarray data deposited in the Gene Expression Omnibus (GEO Profiles, GDS2052). d SCARA5 and DIO2 transcriptlevels quantified by RT-qPCR analysis in 250 endometrial biopsies obtained between LH+ 6 and LH+ 11. Centile calculations were performed on dCtvalues using R software and graphs were generated based on the distribution of expression on each day. The median number of samples for each day was43 (range: 30–46). e Multiplexed single-molecule in situ hybridization (smISH) on three endometrial biopsies (LH+ 10) deemed SCARA5HIGH/DIO2LOW

(95th and 20th percentile, respectively), SCARA5AVERAGE/DIO2AVERAGE (60th and 52th percentile, respectively), and SCARA5LOW/DIO2HIGH (4th and 91stpercentile, respectively). Insert in the left panel shows hybridization with a negative control probe. Insert in middle panel shows SCARA5+ cells (blue,closed arrows) and DIO2+ cells (pink, open arrows) in close proximity. Scale bar: 100 µM. Original magnification: ×40.




pregnancy. Hence, we created four bins based on the sum ofSCARA5 and uNK percentiles and determined the number ofRPL and control subjects in each bin. As shown in Fig. 6b, 81%(25/31) of subjects assigned to the lowest bin were RPL patientswhereas 79% (15/19) in the highest bin were control subjects (P <0.0001; Fisher’s exact test). Our analysis also enabled a definitionof putative defects across the pathway, including ‘decidualizationfailure’ (SCARA5: ≤30th percentile, DIO2: ≤30th percentile),‘excessive decidual senescence’ (SCARA5: ≤30th percentile, DIO2:≥70th percentile), and ‘uNK cell deficiency’ (uNK cells: ≤30thpercentile, SCARA5 and DIO2: >30th−<70th percentile) (Fig. 6c).Excessive decidual senescence and uNK-cell deficiency were sig-nificantly more common in RPL patients compared to controlsubjects (P= 0.0016 and P= 0.02, respectively; χ2 test). Taken

together, the incidence of an aberrant decidual response (i.e.encompassing all defects) was approximately three times higherin RPL compared to control subjects (44% versus 14%, respec-tively, P= 0.00013; Fisher’s exact test).

DiscussionIn menstruating species, including humans, decidualization is notunder the control of an implanting embryo but initiated duringthe midluteal phase of each cycle53. Once triggered, the decid-ualizing endometrium becomes inextricably dependent on sus-tained progesterone signalling. In the absence of implantation,falling progesterone levels elicit an inflammatory decidualresponse which, upon recruitment and activation of leukocytes,

Table 1 Subject demographics.

Controla (n= 90) RPLb (n= 89) P-valuec

Age (years) (median ± IQR) 36 (33–37) 36 (33–38) 0.253BMI (median ± IQR) 22 (21–25) 26 (22–30) <0.0001LH+ day (median ± IQR) 9 (7–10) 9 (7–10) 0.968First trimester loss [median (range)] 0 (0–2) 5 (3–18) <0.0001Live births [mean (range)] 0 (0–2) 0.35 (0–2) <0.0001

aControl subjects were women awaiting IVF treatment for a variety of reasons, including male-factor, unexplained, and tubo-ovarian infertility. All subjects had regular cycles and considered to have goodprognosis; recurrent IVF failure (RIF: >3 consecutive IVF failures with good quality embryos) patients were excludedbAll Recurrent pregnancy loss (RPL) patients in this cohort had three or more consecutive miscarriagescData were tested for normality using Shapiro–Wilk test. P-value was calculated by two-tailed unpaired student’s t-test for normally distributed data (age) or two-tailed Mann Whitney test for non-normally distributed data (BMI, LH+, first trimester loss and live births)

uNKSCARA5 DIO2

0

25

50

75

100

RPL Control RPL

Perc

entil

e

Control RPLControl

b

c

0.00

0.25

0.50

0.75

1.00

(0,50] (50,100] (100,150] (150,200]

SCARA5 and uNK (Bins of sum of percentiles)

Pro

porti

on

RPL Control

P = 1.10 × 10-06P = 2.03 × 10-03 P = 2.47 × 10-02

a25 36 24 4

6 31 38 15

Decidual cells

Senescentdecidual cells

uNK cells

Undifferentiated cells

x x xDecidualization failure

Control: 3/90RPL: 5/89(P=0.46)

Excessive decidual senescenceControl: 6/90RPL: 21/89(P=0.0016)

uNK cell deficiencyControl: 4/90RPL: 13/89

(P=0.02)

Stre

ss

High

Low

uNK cell recruitmentsnDC clearanceSecondary senescence

Fig. 6 Impaired fate divergence of decidual cells in RPL. a Distribution of uNK cells and SCARA5 and DIO2 percentiles in timed endometrial biopsies ofcontrol subjects (n= 90) and RPL patients (n= 89) (Welch two-sided t-test). b Number of RPL and control subjects across four bins defined by the sum ofSCARA5 and uNK-cell percentiles. c Diagram illustrating the decidual pathway. Fate divergence of EnSC upon decidualization relates to the level ofreplicative stress (indicating by nuclear shading) incurred by individual cells during the proliferative phase. Stress-resistant decidual cells recruit andactivate uNK cells to eliminate stressed/senescent decidual cells through granule exocytosis. Different defects along the decidual pathway can beidentified, including ‘decidualization failure’, ‘excessive decidual senescence’ and ‘uNK cell deficiency’. The frequency of each defect in RPL and controlsubjects is shown (χ2 test).





leads to tissue breakdown and menstrual shedding54,55. Thus, acritical challenge at implantation is to simultaneously avoidimminent endometrial breakdown while transforming the cyclingendometrium into a semi-permanent tissue, the decidua, main-tained throughout pregnancy. We recently argued that successfultransformation of the endometrium in early pregnancy requiresboth successful differentiation of EnSC into DC and promptelimination of snDC by activated uNK cells; and conversely, thatdefects in this process predisposes for early pregnancy loss9. Totest our hypothesis, we set out to identify putative marker genesof an aberrant decidual response using single-cell RNA-seq.

We first generated a detailed transcriptional map of primaryEnSC decidualized in culture over 8 days followed by withdrawalof the differentiation signal for 2 days. This analysis revealed amultistep process that starts with a precipitous transcriptionalresponse in differentiating EnSC, which is followed by synchro-nous transition of cells through intermediate states and then theemergence of DC and some snDC. A recent study asserted thatDC emerged in evolution from rewiring of an ancestral cellularstress response19. This conclusion was based on the observationthat endometrial fibroblasts isolated from marsupials, whichdiverted from eutherians 60 to 80 million years ago56, activatesimilar core regulatory genes as human EnSC in response to adecidualizing stimulus and then mount an acute stress responseakin to the initial decidual phase19. The key difference is thatmost EnSC emerge from this process as DC expressing multipleanti-inflammatory and stress-resistance genes. However, someEnSC emerge as snDC, thus recapitulating the ancestral response.We also demonstrated that snDC rapidly convert DC into sec-ondary senescent cells, a process abrogated by co-cultured uNKcells. Importantly, secondary senescence had a dramatic impacton decidual gene expression, characterized by marked upregula-tion of genes encoding for ECM proteins and proteases and otherSASP components. The mechanisms driving secondary senes-cence of DC warrant further exploration, including the possibilitythat EnSC become sensitized to juxtracrine senescence upondifferentiation into DC. A hallmark of snDC is progesterone-resistance, exemplified by expression of PGR-repressed genes,such as SOX4 and FOXP1,14 and lack of responsiveness to pro-gesterone and cAMP withdrawal. Analysis of co-regulated genenetworks indicated that a switch in PGR co-regulators fromFOSL2 to STAT1 may account for progesterone-resistance insnDC, although this conjecture will need to be tested experi-mentally. Apart from generating a temporal map of TF andeffector genes underpinning progression of cells along thedecidual pathway, the network analysis also revealed that multiplegenes involved in immune surveillance of senescent cells,including CXCL14, IL-15 and TIMP-3, are hardwired to be acti-vated prior to the emergence of snDC. Interestingly, humanchorionic gonadotrophin has been shown to stimulate uNK-cellproliferation directly57, which suggests potential cooperationbetween the implanting embryo and DC in eliminating snDC.

Our in vitro analysis suggested that the default trajectory ofdecidualizing EnSC is towards cellular senescence, which can onlybe avoided by timely clearance of snDC by uNK cells. Failure todo so leads to chronic senescence, at least in culture. There aremultiple mechanisms by which chronic senescent cells promotetissue dysfunction, including perturbation of stem cells, disrup-tion of ECM, induction of tissue inflammation, and propagationof senescence in neighbouring cells25,27,32. When extrapolated tothe in vivo situation, these observations imply that a pro-senescent decidual response during the peri-implantation windowwould lead to the formation of an intrinsically unstabledecidual–placental interface in pregnancy. To test this hypothesis,we set out to identify stroma-specific marker genes to dis-criminate between a physiological and pathological decidual

response. Single-cell transcriptomic analysis of midluteal endo-metrium yielded a number of notable results, including the dis-covery of a discrete but as yet uncharacterized population ofproliferative mesenchymal cells, the characterization of differentepithelial cell subsets, and the identification of three uNK-cellstates defined in part by the abundance of cell-cycle genes. Forexample, the NK1 population, representing the most proliferatinguNK cells, abundantly express genes involved in granule exocy-tosis (e.g. PRF1, GNLY, GZMA and GZMB). By contrast, NK3cells express low levels of cell-cycle genes but are defined by CCL5and CXCR4 expression. Notably, CXCR4+ uNK cells have pre-viously been implicated in vascular remodelling in pregnancy58.

As the in vivo single-cell analysis was restricted to few samplesacross two time-points, we restricted our search for putativemarker genes to the 50 top in vitro branching genes that markeddivergence of DC and DC already transitioning towards senes-cence. Out of 50 genes, SCARA5 and DIO2 were deemed infor-mative marker genes of DC and snDC, respectively, based ontheir temporal regulation across the cycle in whole endometrialbiopsies, lack of regulation in glandular epithelium, and level ofenrichment in stromal cells. Further, multiplexed smISH showedthat SCARA5+ cells are distinct from DIO2+ cells, which wasfurther corroborated by the different transcriptomic profiles ofSCARA5enriched/DIO2reduced and SCARA5reduced/DIO2enriched

EnSC in the in vivo single-cell data set. To test our suppositionthat an aberrant decidual trajectory predisposes to pregnancy loss,we first generated percentile graphs for SCARA5 and DIO2expression using 250 midluteal biopsies (LH+ 6–11). Thisapproach enables comparison of the relative level of geneexpression in biopsies obtained at different days of the cycle. Therelative abundance of uNK cells was also determined using apreviously established percentile graph based on analysis of 1997biopsies9. Next, we quantified the abundance of uNK cells and theexpression of SCARA5 and DIO2 in timed endometrial biopsiesfrom RPL and control subjects. Sample selection was based solelyon reproductive history; all RPL patients had three or moreconsecutive miscarriages whereas control subjects were womenwith male-factor, tubal or unexplained infertility awaiting IVFtreatment. Overall, our analysis indicated that pre-pregnancyendometrium in RPL is characterized by uNK-cell deficiency inparallel with a shift from a preponderance of DC to snDC. Next,we examined the frequency of different putative defects along thedecidual pathway. Based on pre-specified but arbitrary criteria, wefound that RPL is associated with ‘excessive decidual senescence’and ‘uNK cell deficiency’ but not ‘decidualization failure’. Whilepromising, prospective studies are needed to define prognosticcriteria that could be exploited for pre-pregnancy screening ofwomen at risk of RPL.

Decidualization is an iterative process in cycling endometrium.At present, it is not clear if an aberrant decidual response persistsfrom cycle-to-cycle or whether it represents an intermittent defectthat affects some but not all cycles. The relative high cumulativelive-birth rate in RPL favours the latter scenario59. Two distinctregulatory mechanisms control cellular homoeostasis in cyclingendometrium. On the one hand, uNK cells may engender selec-tive elimination of snDC during the midluteal phase, de factorejuvenating the endometrium at the time of embryo implanta-tion9. On the other, the cycling endometrium actively recruitsbone marrow-derived cells (BDMC) capable of differentiatinginto stromal, epithelial and endothelial cells60. Experimentalstudies in mice have shown that BDMC enhance the regenerativecapacity of non-pregnant endometrium60, and contribute to rapiddecidual expansion in pregnancy61. Interestingly, we recentlyreported a double-blind, randomised, placebo-controlled pilottrial demonstrating that sitagliptin, an oral antidiabetic drug andDPP4 inhibitor, is effective in increasing engraftment of BMDC




in the endometrium and attenuating the pro-senescent decidualresponse in RPL patients62.

In summary, based on single-cell transcriptomic analysis, wepropose that decidualization is a multistep process that ultimatelyleads to chronic senescence, a cellular state incompatible with theformation of a functional decidual–placental interface. Our datasuggest that DC must engage uNK cells to eliminate snDC toescape this default pathway, although additional in vivo studiesare required to substantiate this conjecture. Nevertheless, weidentified SCARA5 and DIO2 as selective marker genes for DCand progesterone-resistant snDC, respectively. When combinedwith uNK cells, these marker genes can be used to map putativedefects along the decidual pathway in cycling endometrium. Ourfindings raise the possibility of a simple screening test to identifywomen at risk of miscarriage and to monitor the effectiveness ofpre-pregnancy interventions.

MethodsEthical approval and sample collection. The study was approved by the NHSNational Research Ethics-Hammersmith and Queen Charlotte’s & ChelseaResearch Ethics Committee (1997/5065). All samples were obtained with writteninformed consent and in accordance with The Declaration of Helsinki (2000)guidelines. Human endometrial biopsies were obtained from women attending theImplantation Clinic, a dedicated research clinic at University Hospitals Coventryand Warwickshire (UHCW) National Health Service Trust. Surplus tissue fromendometrial biopsies obtained for diagnostic purposes at the ImplantationResearch Clinic was used for this study. Samples were obtained during the lutealphase of ovulatory, non-hormonally stimulated menstrual cycles, timed relative tothe pre-ovulatory LH surge, using a Wallach Endocell™ endometrial sampler. Overtuterine pathology was excluded by transvaginal ultrasound scan prior to the biopsy.Control subjects were women awaiting IVF treatment for a variety of reasons,including male-factor, unexplained, and tubo-ovarian infertility. Control subjectswere considered to have good prognosis; patients with recurrent implantationfailure, defined as three or more consecutive IVF failures with good qualityembryos, were excluded. Recurrent pregnancy loss (RPL) patients had three ormore consecutive miscarriages. Demographic details are presented in Table 1.

Primary endometrial stromal cell (EnSC) culture. Endometrial biopsies werecollected in DMEM-F12 media supplemented with 10% dextran-coated charcoal-stripped FBS (DCC) and processed for primary EnSC culture as described9. Briefly,tissue was minced for 5 min before digestion with Collagenase Type Ia (500 µg/ml)and DNase I (100 µg/ml) for 1 h at 37 °C. Cell suspension was then filtered througha 40 µM cell sieve to remove undigested glandular cell clumps. The flow-throughwas cultured in DMEM/F12 with phenol red containing 10% DCC, 1X Antibiotic-Antimycotic mix, L-glutamine, estradiol and insulin (growth medium). Cells werecultured on tissue culture-treated plastic with media changed every 48 h. Cells werepassaged at sub-confluence by 5 min treatment with 0.25% Trypsin-EDTA. Fordecidualization studies, confluent monolayers of human endometrial stromal cells(EnSC) at passage 2 were incubated overnight at 37 °C with 5% CO2 in phenol red-free DMEM/F-12 containing 2% DCC, antibiotic/antimycotic and L-glutamine (2%media). To induce differentiation, cells were treated with 0.5 mM 8-bromo-cAMP(Sigma-Aldrich, Poole, UK) and 1 μM medroxyprogesterone acetate (MPA; Sigma-Aldrich) for the indicated time-points. For inhibition of senescence, cells weretreated with 250 nM dasatinib (CST, Leiden, The Netherlands) throughout thedifferentiation time-course.

Droplet generation and single-cell sequencing (Drop-Seq). Single-cell tran-scriptomes were captured in aqueous droplets containing barcoded beads using amicrofluidic system (scRNA-seq: Dolomite Bio, Royston, UK) according to themanufacturer’s protocol and based on the Drop-Seq method described by Macoskoand colleagues34. Briefly, cells in suspension were placed into the remote chamber ofthe scRNA-seq system. Barcoded beads (Barcoded Bead SeqB; Chemgenes Corp.,USA) in lysis buffer at a concentration of 280 beads/µl were loaded into the sampleloop. Cell and bead solutions were run at a flow rate of 30 µl/min into a fluorophilicglass microfluidic chip with 100 µm etch depth (Single Cell RNA-seq Chip, Dolo-mite Bio) with droplet generation oil (Bio-Rad Laboratories, UK) at a flow rate of200 µl/min for 15–18min. Droplets were collected into a 50ml Falcon tube, qualitychecked using a C-Chip Fuchs-Rosenthal Haemocytometer (Labtech, Heathfield,UK), and bead doublets counted. Droplet breakage, bead isolation and reversetranscription were performed exactly as described by Macosko and Goldman(Drop-Seq Laboratory Protocol version 3.1, http://mccarrolllab.org/download/905/)on 8000 beads per reaction, with two reactions per sample, to give ~800 single-celltranscriptomes attached to microparticles (STAMPS) per timepoint. PCR condi-tions were as per Macosko and Goldman. Clean-up with Agencourt AMPure XPbeads (Beckman Coulter, High Wycombe, UK) was performed according to stan-dard Illumina RNA-seq protocols with a 0.6 × beads to sample ratio. cDNA was

eluted in 12 µl nuclease-free water and quality and size assessed using an AgilentBioanalyzer High Sensitivity DNA chip (Agilent, Stockport, UK). For tagmentation600 pg cDNA, as determined by Qubit High Sensitivity DNA assay (ThermoFisher,Paisley, UK), was processed according to Macosko and Goldman using IlluminaNextera XT DNA Sample Kit and Indexing Kit (Illumina, Cambridge, UK). Tag-mented libraries were cleaned up using AMPure XP beads as before, with a 0.6 ×beads ratio followed by a repeat clean-up using 1× beads. Eluted libraries wereanalysed using Agilent Bioanalyzer High Sensitivity DNA chip to assess quality anddetermine library size, and concentration was determined by Qubit High SensitivityDNA assay. Library dilution and denaturation was performed as per standardIllumina protocols and sequenced using NextSeq High Output 75 cycle V2 kit.

Drop-Seq data alignment and quantification. Initial Drop-Seq data processingwas performed using Drop-Seq_tools-1.0.1 following the protocol described byNemesh (seqAlignmentCookbook_v1.2Jan2016.pdf, http://mccarrolllab.com/dropseq). Briefly, reads with low-quality bases in either cell or molecular barcodewere filtered and trimmed for contaminating primer or poly-A sequence.Sequencing errors in barcodes were inferred and corrected, as implemented byDrop-Seq_tools-1.0.1. Reads were aligned to the hg19 (Human) reference genomeconcatenated with ERCC annotations using STAR-2.5.3a63, with the Gencode21(Human) as reference transcriptome. Uniquely mapped reads, with ≤1 insertion ordeletion, were used in quantification. Finally, the DigitalExpression tool34 was usedto obtain the digital gene expression matrix for each sample. Cell numbers wereselected computationally from the inflection point in a cumulative distribution ofreads plotted against the cell barcodes ordered by descending number of reads64.Cell barcodes beyond the inflection point are believed to represent ‘ambient RNA’(e.g. contaminating RNA from damaged cells), not cellular transcriptomes, andtherefore excluded from further analysis. This resulted in ~800 cells per timepoint,matching the number anticipated from processed bead counts.

Cell aggregation analysis. Analysis of DGE data was performed with Seurat v265.To select high-quality data for analysis, cells were included when at least 200 geneswere detected, while genes were included if they were detected in at least three cells.Cells which had more than 4500 genes were excluded from the analysis as werecells with more than 5% mitochondrial gene transcripts to minimize doublets andlow-quality (broken or damaged) cells, respectively. After scaling and normal-ization of the raw counts in the DGE matrix, cell-cycle regression was applied. Forcell aggregation, a set of highly variable genes was first identified, with an averageexpression mean between 0.0125 and 3 and a Log Variant to Mean Ratio of at least0.5, which were used to perform principal component (PC) analysis. Based onstatistical significance and the robustness of the results, the first 10 PCs weresubsequently used as inputs for clustering via shared nearest neighbour (SNN) andsubsequent t-distributed stochastic neighbour embedding (t-SNE) representation.The Seurat function ‘FindAllMarkers’ employing the Wilcoxon test was used toidentify marker genes for each cell state cluster in the t-SNE representation. Toobtain independent estimation of the number of unique cell types we used SC3v1.3.18 (5), applying a consensus strategy and Tracy-Widom theory on randommatrices to estimate the optimal number of clusters (k).

Co-regulated gene networks. K-means cluster analysis was performed in MeVv4.8 (6) to group marker genes based on co-expression across cell state clusters.Figure-of-merit (FOM) was run first to determine the number of expression pat-terns (k). The predictive power of the k-means algorithm was estimated using anFOM values for k from 1 to 20. K-means clustering was run using Pearson cor-relation metric for a maximum of 50 iterations. Gene ontology analysis was per-formed on the clustered genes using DAVID66.

Trajectory analysis. Trajectory analysis was performed with Monocle v2.6.167.Branch-point genes were identified with Branched Expression Analysis Modeling(BEAM) function.

Gene expression correlation. To test gene expression correlation between pairsof genes, expression was imputed for every cell using Markov Affinity-basedGraph Imputation of Cells (MAGIC)68. Pearson correlation coefficients werecalculated for top 50 genes determining the different trajectories at informativebranch-points. To assess congruency between the time-course and biopsy datasets,the correlation coefficients were added resulting in sum of coefficients between−2 and +2. To infer gene expression associated with a pro-senescent decidualresponse, SCARA5enriched/DIO2reduced and SCARA5reduced/DIO2enriched EnSCswere selected after MAGIC imputation using the midpoint of the expression levelsas thresholds. Differential gene expression on the selected cells was determinedusing the Seurat function ‘FindMarkers’ employing the Wilcoxon test, with GOanalysis performed using DAVID66.

Drop-Seq analysis of timed endometrial biopsies. Six LH-timed endometrialbiopsies were processed as described in detail above and elsewhere9. Aftertissue digestion, red blood cells were removed from the flow-through by Ficolldensity gradient centrifugation (ref. 9). Single-cell fractions were then subjected to



http://mccarrolllab.org/download/905/

http://mccarrolllab.com/dropseq

http://mccarrolllab.com/dropseq



Drop-Seq analysis. The time from biopsy in the clinic to cDNA synthesis was <4 hfor all samples. Anonymized endometrial biopsies were obtained from women agedbetween 31 and 42 years with regular cycles, body mass index between 23 and 32 kg/m2,and absence of uterine pathology on transvaginal ultrasound examination.

Reverse transcription quantitative PCR (RT-qPCR). RNA was extracted fromendometrial biopsies which had been snap frozen in clinic (<1 min after collection),using STAT-60 (AMS Biotechnology, Oxford, UK) according to the manufacturer’sinstructions. Reverse transcription was performed from 1 µg RNA using theQuantitect Reverse Transcription Kit (QIAGEN, Manchester, UK) and cDNA wasdiluted to 10 ng/µl equivalent before use in qPCR. Amplification was performed ona 7500 Real-Time PCR system (Applied Biosystems, Paisley, UK) in 20 µl reactionsusing 2 × PrecisionPlus Mastermix with SYBR Green and low ROX (PrimerDesign,Southampton, UK), with 300 nM each of forward and reverse primers. L19 wasused as a reference control. Primer sequences were as follows: SCARA5 forward:5′-CATGCG TGG GTT CAA AGG TG-3′, SCARA5 reverse: 5′-CCA TTC ACCAGG CGG ATC AT-3′; DIO2 forward: 5′-ACT CGG TCA TTC TGC TCA A-3′,DIO2 reverse: 5′-TTC CAG ACG CAG CGC AGT-3′, L19 forward: 5′-GCG GAAGGG TAC AGC CAA T-3′, L19 reverse: 5′-GCA GCC GGC GCA AA-3′. Centilecalculations were performed on dCt values using R v3.5 software.

Multiplexed single-molecule in situ hybridization. Formalin-fixed paraffin-embedded (FFPE) samples were cut to 5 µm sections. RNA in situ hybridizationwas carried with RNAscope® 2.5 HD Duplex Reagent Kit (ACD, California, USA)with probes for SCARA5 (574781-C1) and DIO2 (562211-C2) according tomanufacturer’s guidelines. Following hybridization and amplification, slides werecounterstained with 50% haematoxylin. Images were obtained using an EVOSAUTO microscope (ThermoFisher Scientific) with a ×40 objective lens.

Isolation and culture of uNK cells. Primary uNK cells were isolated from lutealphase endometrial biopsies as described previously9. Briefly, supernatant fromfreshly digested EnSC cultures was collected uNK cells were isolated by magnetic-activated cell separation (MACS; Miltenyi Biotec, Bergisch Gladbach, Germany)using phycoerythrin (PE)-conjugated anti-CD56 antibody (Bio-Legend, San Diego,CA, USA), as per manufacturer’s instructions. The CD56+ positive fraction wascollected by centrifugation and cultured in suspension for up to 5 days in RPMImedia (Sigma-Aldrich) supplemented with 10% DCC-FBS, 1× Antibiotic-Anti-mycotic, and 2 ng/ml IL-15 (Sigma-Aldrich) to aid uNK-cell maturation. Toincrease yield, uNK cells from 3 to 5 subjects were pooled. For co-cultureexperiments, uNK cells were pelleted and re-suspended in DMEM/F-12 containing2% DCC without IL-15. A total of 5000 uNK cells were added to 50,000 EnSC,decidualized or not, per well of a 96-well plate. MACS isolated uNK cells wereanalysed by flow cytometry after labelling with BD Horizon™ Fixable Viability Stain(FVS) 660 (BD Biosciences) in 1 × PBS (1:1000) and PE-conjugated CD56 (BDBiosciences; clone: B159; catalogue no: 555516; 1:5) according to manufacturer’sinstructions. Red blood cells (RBC) were removed from the negative fraction using1×RBC lysis solution (BD Biosciences, catalogue number: 349202) for 10 min atroom temperature. Samples were analysed by BD FACSMelody™ Cell Sorter (BDBiosciences) and data analysis was performed using FlowJo v10.

Quantitative analysis of uNK cells in endometrial biopsies. FFPE tissue sectionswere stained for CD56 (a uNK-cell-specific cell surface antigen) using a 1:200dilution of concentrated CD56 antibody (NCL-L-CD56-504, Novocastra, LeicaBioSystems). Stained slides were de-hydrated, cleared and cover-slipped in aTissue-Tek® Prisma® Automated Slide Stainer, model 6134 (Sakura Flinetek Inc.,CA, USA) using DPX coverslip mountant. Bright-field images were obtained on aMirax Midi slide scanner using a ×20 objective lens and opened in PanoramicViewer v1.15.4 (3DHISTECH Ltd, Budapest, Hungary) for analysis. To avoidinconsistencies that reflect reduced uNK-cell densities at greater endometrialdepths, CD56+ cells were quantified in compartments directly underlying theluminal epithelium. Here, three randomly selected areas of interest for each biopsywere identified and captured within Panoramic Viewer before analysis in ImageJimage analysis software. Both luminal and glandular epithelial cells were removedmanually before colour deconvolution into constituent brown (CD56+ staining)and blue (hematoxylin staining – stromal cells). The area of positive staining abovea manually determined background threshold was used to quantify stainingintensity. The uNK-cell percentage was calculated as the number of CD56+ cellsper 100 stromal cells and averaged from three images9.

Enzyme-linked immunosorbent assay (ELISA). Culture supernatants were col-lected every 2 days and centrifuged to clear cell debris prior to storage at −20 °C.Analytes in collected supernatant were measured by ELISA as per manufacturer’sinstructions (DuoSet ELISA kits, Bio-Techne, Abingdon, UK). Standard curveswere fitted to a 4-parameter logistic fit curve in GraphPad Prism v8 software andsample concentrations interpolated from these graphs.

Quantitation of SAβG activity. SAβG activity in cultured cells was quantifiedusing the 96-Well Cellular Senescence Activity Assay kit (CBA-231, Cell BiolabsInc, CA, USA) as described previously9.

Statistical analysis and reproducibility. Chi-square and Fisher’s exact test on thedecidual pathway defects were performed using online calculators at https://www.socscistatistics.com/. All other analyses were implemented in R. Single-cell RNA-seq was validated on three independent human primary endometrial stromal cellcultures, and six endometrial biopsies in vivo. A seventh sample was excluded fromthe in vivo analysis due to a considerable batch effect introduced by processing ofthis sample in isolation from the other six. A minimum of four biological repeatswere used for secretome analyses. In addition, 250 timed endometrial biopsies wereused to generate the percentile graphs, and 90 and 89 biopsies of control subjectsand recurrent pregnancy loss patients, respectively, were analysed. The sample sizeof clinical samples was determined by the availability of biopsy samples frompatients with the appropriate phenotype.

Reporting summary. Further information on research design is available inthe Nature Research Reporting Summary linked to this article.

Data availabilityThe Drop-Seq data were deposited in the GEO repository (http://www.ncbi.nlm.nih.gov/geo/with accession number: GSE127918). All other data are available on request to thecorresponding author.

Received: 29 April 2019; Accepted: 2 January 2020;

References1. Rai, R. & Regan, L. Recurrent miscarriage. Lancet 368, 601–611 (2006).2. Hardy, K., Hardy, P. J., Jacobs, P. A., Lewallen, K. & Hassold, T. J. Temporal

changes in chromosome abnormalities in human spontaneous abortions:results of 40 years of analysis. Am. J. Med Genet A 170, 2671–2680 (2016).

3. Carp, H. et al. Karyotype of the abortus in recurrent miscarriage. Fertil. Steril.75, 678–682 (2001).

4. Stephenson, M. D., Awartani, K. A. & Robinson, W. P. Cytogenetic analysis ofmiscarriages from couples with recurrent miscarriage: a case-control study.Hum. Reprod. 17, 446–451 (2002).

5. ESHRE. Recurrent Pregnancy Loss: A Guideline of the European Society ofHuman Reproduction and Embryology (2017).

6. Practice Committee of the American Society for Reproductive, M. Evaluationand treatment of recurrent pregnancy loss. Fertil. Steril. 5, 1103–1111 (2012).

7. Ogasawara, M., Aoki, K., Okada, S. & Suzumori, K. Embryonic karyotype ofabortuses in relation to the number of previous miscarriages. Fertil. Steril. 73,300–304 (2000).

8. Gellersen, B. & Brosens, J. J. Cyclic decidualization of the humanendometrium in reproductive health and failure. Endocr. Rev. 35, 851–905(2014).

9. Brighton, P. J. et al. Clearance of senescent decidual cells by uterine naturalkiller cells in cycling human endometrium. eLife 6, e31274 (2017).

10. Weimar, C. H. et al. Endometrial stromal cells of women with recurrentmiscarriage fail to discriminate between high- and low-quality humanembryos. PLoS ONE 7, e41424 (2012).

11. Brosens, J. J. et al. Uterine selection of human embryos at implantation. Sci.Rep. 4, 3894 (2014).

12. Brosens, J. J., Pijnenborg, R. & Brosens, I. A. The myometrial junctional zonespiral arteries in normal and abnormal pregnancies: a review of the literature.Am. J. Obstet. Gynecol. 187, 1416–1423 (2002).

13. Vrljicak, P. et al. Analysis of chromatin accessibility in decidualizing humanendometrial stromal cells. FASEB J. 32, 2467–2477 (2018).

14. Cloke, B. et al. The androgen and progesterone receptors regulate distinct genenetworks and cellular functions in decidualizing endometrium. Endocrinology149, 4462–4474 (2008).

15. Leitao, B. et al. Silencing of the JNK pathway maintains progesterone receptoractivity in decidualizing human endometrial stromal cells exposed to oxidativestress signals. FASEB J. 24, 1541–1551 (2010).

16. Muter, J. et al. Progesterone-dependent induction of phospholipase C-relatedcatalytically inactive protein 1 (PRIP-1) in decidualizing human endometrialstromal cells. Endocrinology 157, 2883–2893 (2016).

17. Lynch, V. J., Leclerc, R. D., May, G. & Wagner, G. P. Transposon-mediatedrewiring of gene regulatory networks contributed to the evolution ofpregnancy in mammals. Nat. Genet. 43, 1154–1159 (2011).

18. Lynch, V. J. et al. Ancient transposable elements transformed the uterineregulatory landscape and transcriptome during the evolution of mammalianpregnancy. Cell Rep. 10, 551–561 (2015).



https://www.socscistatistics.com/

https://www.socscistatistics.com/

http://www.ncbi.nlm.nih.gov/geo/


19. Erkenbrack, E. M. et al. The mammalian decidual cell evolved from a cellularstress response. PLoS Biol. 16, e2005594 (2018).

20. Al-Sabbagh, M. et al. NADPH oxidase-derived reactive oxygen speciesmediate decidualization of human endometrial stromal cells in response tocyclic AMP signaling. Endocrinology 152, 730–740 (2011).

21. Salker, M. S. et al. Disordered IL-33/ST2 activation in decidualizing stromalcells prolongs uterine receptivity in women with recurrent pregnancy loss.PLoS ONE 7, e52252 (2012).

22. Kajihara, T. et al. Differential expression of FOXO1 and FOXO3a confersresistance to oxidative cell death upon endometrial decidualization. Mol.Endocrinol. 20, 2444–2455 (2006).

23. Kuroda, K. et al. Induction of 11beta-HSD 1 and activation of distinctmineralocorticoid receptor- and glucocorticoid receptor-dependent genenetworks in decidualizing human endometrial stromal cells. Mol. Endocrinol.27, 192–202 (2013).

24. Antonangeli, F., Zingoni, A., Soriani, A. & Santoni, A. Senescent cells: living ordying is a matter of NK cells. J. Leukoc. Biol. 105, 1275–1283 (2019).

25. Childs, B. G. et al. Senescent cells: an emerging target for diseases of ageing.Nat. Rev. Drug Discov. 16, 718 (2017).

26. Hernandez-Segura, A., Nehme, J. & Demaria, M. Hallmarks of cellularsenescence. Trends Cell Biol. 28, 436–453 (2018).

27. van Deursen, J. M. The role of senescent cells in ageing. Nature 509, 439–446(2014).

28. Jun, J. I. & Lau, L. F. The matricellular protein CCN1 induces fibroblastsenescence and restricts fibrosis in cutaneous wound healing. Nat. Cell Biol.12, 676–685 (2010).

29. Storer, M. et al. Senescence is a developmental mechanism that contributes toembryonic growth and patterning. Cell 155, 1119–1130 (2013).

30. Ritschka, B. et al. The senescence-associated secretory phenotype inducescellular plasticity and tissue regeneration. Genes Dev. 31, 172–183 (2017).

31. Acosta, J. C. et al. Chemokine signaling via the CXCR2 receptor reinforcessenescence. Cell 133, 1006–1018 (2008).

32. Ito, Y., Hoare, M. & Narita, M. Spatial and temporal control of senescence.Trends Cell Biol. 27, 820–832 (2017).

33. Lucas, E. S. et al. Loss of endometrial plasticity in recurrent pregnancy loss.Stem Cells 34, 346–356 (2016).

34. Macosko, E. Z. et al. Highly parallel genome-wide expression profiling ofindividual cells using nanoliter droplets. Cell 161, 1202–1214 (2015).

35. Kuroda, K. et al. Elevated periimplantation uterine natural killer cell density inhuman endometrium is associated with impaired corticosteroid signaling indecidualizing stromal cells. J. Clin. Endocrinol. Metab. 98, 4429–4437 (2013).

36. Song, J. J. et al. Role of glutaredoxin in metabolic oxidative stress.Glutaredoxin as a sensor of oxidative stress mediated by H2O2. J. Biol. Chem.277, 46566–46575 (2002).

37. Zuo, R. J. et al. Crystallin alphaB acts as a molecular guard in mousedecidualization: regulation and function during early pregnancy. FEBS Lett.588, 2944–2951 (2014).

38. Latini, F. R. et al. ABI3 ectopic expression reduces in vitro and in vivo cellgrowth properties while inducing senescence. BMC Cancer 11, 11 (2011).

39. Michishita, E., Garces, G., Barrett, J. C. & Horikawa, I. Upregulation of theKIAA1199 gene is associated with cellular mortality. Cancer Lett. 239, 71–77(2006).

40. Petropoulou, C., Trougakos, I. P., Kolettas, E., Toussaint, O. & Gonos, E. S.Clusterin/apolipoprotein J is a novel biomarker of cellular senescence thatdoes not affect the proliferative capacity of human diploid fibroblasts. FEBSLett. 509, 287–297 (2001).

41. Trougakos, I. P. The molecular chaperone apolipoprotein J/clusterin as asensor of oxidative stress: implications in therapeutic approaches—a mini-review. Gerontology 59, 514–523 (2013).

42. Bianco, A. C. & Kim, B. W. Deiodinases: implications of the local control ofthyroid hormone action. J. Clin. Invest 116, 2571–2579 (2006).

43. Severino, V. et al. Insulin-like growth factor binding proteins 4 and 7 releasedby senescent cells promote premature senescence in mesenchymal stem cells.Cell Death Dis. 4, e911 (2013).

44. Le Maitre, C. L., Freemont, A. J. & Hoyland, J. A. Accelerated cellularsenescence in degenerate intervertebral discs: a possible role in thepathogenesis of intervertebral disc degeneration. Arthritis Res. Ther. 9, R45(2007).

45. Brosens, J. J., Hayashi, N. & White, J. O. Progesterone receptor regulatesdecidual prolactin expression in differentiating human endometrial stromalcells. Endocrinology 140, 4809–4820 (1999).

46. Mazur, E. C. et al. Progesterone receptor transcriptome and cistrome indecidualized human endometrial stromal cells. Endocrinology 156, 2239–2253(2015).

47. Christian, M. et al. Interferon-gamma modulates prolactin and tissue factorexpression in differentiating human endometrial stromal cells. Endocrinology142, 3142–3151 (2001).

48. Mokhtar, N. M. et al. Progestin regulates chemokine (C-X-C motif) ligand 14transcript level in human endometrium. Mol. Hum. Reprod. 16, 170–177(2010).

49. Raneros, A. B. et al. Increasing TIMP3 expression by hypomethylating agentsdiminishes soluble MICA, MICB and ULBP2 shedding in acute myeloidleukemia, facilitating NK cell-mediated immune recognition. Oncotarget 8,31959–31976 (2017).

50. Altmae, S. et al. Meta-signature of human endometrial receptivity: a meta-analysis and validation study of transcriptomic biomarkers. Sci. Rep. 7, 10077(2017).

51. Vento-Tormo, R. et al. Single-cell reconstruction of the early maternal-fetalinterface in humans. Nature 563, 347–353 (2018).

52. Lash, G. E. et al. Standardisation of uterine natural killer (uNK) cellmeasurements in the endometrium of women with recurrent reproductivefailure. J. Reprod. Immunol. 116, 50–59 (2016).

53. Emera, D., Romero, R. & Wagner, G. The evolution of menstruation: a newmodel for genetic assimilation: explaining molecular origins of maternalresponses to fetal invasiveness. Bioessays 34, 26–35 (2012).

54. Evans, J. & Salamonsen, L. A. Inflammation, leukocytes and menstruation.Rev. Endocr. Metab. Disord. 13, 277–288 (2012).

55. Evans, J. & Salamonsen, L. A. Decidualized human endometrial stromal cellsare sensors of hormone withdrawal in the menstrual inflammatory cascade.Biol. Reprod. 90, 14 (2014).

56. O’Leary, M. A. et al. The placental mammal ancestor and the post-K-Pgradiation of placentals. Science 339, 662–667 (2013).

57. Kane, N., Kelly, R., Saunders, P. T. & Critchley, H. O. Proliferation of uterinenatural killer cells is induced by human chorionic gonadotropin and mediatedvia the mannose receptor. Endocrinology 150, 2882–2888 (2009).

58. Gibson, D. A., Greaves, E., Critchley, H. O. & Saunders, P. T. Estrogen-dependent regulation of human uterine natural killer cells promotes vascularremodelling via secretion of CCL2. Hum. Reprod. 30, 1290–1301 (2015).

59. Ewington, L. J., Tewary, S. & Brosens, J. J. New insights into the mechanismsunderlying recurrent pregnancy loss. J. Obstet. Gynaecol. Res. 45, 258–265(2019).

60. Santamaria, X., Mas, A., Cervello, I., Taylor, H. & Simon, C. Uterine stemcells: from basic research to advanced cell therapies. Hum. Reprod. Update 24,673–693 (2018).

61. Tal, R. et al. Adult bone marrow progenitors become decidual cells andcontribute to embryo implantation and pregnancy. PLoS Biol. 17, e3000421(2019).

62. Tewary, S. et al. Impact of sitagliptin on endometrial mesenchymal stem-likeprogenitor cells: A randomised, double-blind placebo-controlled feasibilitytrial. EBioMedicine, https://doi.org/10.1016/j.ebiom.2019.102597 (2020).

63. Dobin, A. et al. STAR: ultrafast universal RNA-seq aligner. Bioinformatics 29,15–21 (2013).

64. Lun, A. T. L. et al. EmptyDrops: distinguishing cells from empty droplets indroplet-based single-cell RNA sequencing data. Genome Biol. 20, 63 (2019).

65. Satija, R., Farrell, J. A., Gennert, D., Schier, A. F. & Regev, A. Spatialreconstruction of single-cell gene expression data. Nat. Biotechnol. 33,495–502 (2015).

66. Huang da, W., Sherman, B. T. & Lempicki, R. A. Systematic and integrativeanalysis of large gene lists using DAVID bioinformatics resources. Nat. Protoc.4, 44–57 (2009).

67. Trapnell, C. et al. The dynamics and regulators of cell fate decisions arerevealed by pseudotemporal ordering of single cells. Nat. Biotechnol. 32,381–386 (2014).

68. van Dijk, D. et al. Recovering gene interactions from single-cell data usingdata diffusion. Cell 174, 716–729 (2018).

AcknowledgementsWe are grateful to all the women who participated in this research. This work wassupported by funds from the Tommy’s National Miscarriage Research Centre andWellcome Trust Investigator Award to J.J.B. and S.O. (212233/Z/18/Z).

Author contributionsConceptualization: J.J.B.; methodology: P.V., E.S.L, J.L. and S.O; investigation: E.S.L.,P.V., J.M., M.D.C., P.J.B., C-S.K. and K.F.; writing - original draft: J.J.B., P.V. and E.S.L.;funding acquisition: S.Q., S.O. and J.J.B.; resources: J.O., L.J.E., S.Q. and J.J.B.; super-vision: S.O. and J.J.B.

Competing interestsThe authors declare no competing non-financial interests but the following competingfinancial interests: Patent application, The University of Warwick, Professor Jan Brosens,application no. 1911947.8, pending, identification of SCARA5 and DIO2 as marker genes



https://doi.org/10.1016/j.ebiom.2019.102597



of decidual cells and senescent decidual cells, respectively. The remaining authors declareno competing financial interests.

Additional informationSupplementary information is available for this paper at https://doi.org/10.1038/s42003-020-0763-1.

Correspondence and requests for materials should be addressed to J.J.B.

Reprints and permission information is available at http://www.nature.com/reprints

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims inpublished maps and institutional affiliations.

Open Access This article is licensed under a Creative CommonsAttribution 4.0 International License, which permits use, sharing,

adaptation, distribution and reproduction in any medium or format, as long as you giveappropriate credit to the original author(s) and the source, provide a link to the CreativeCommons license, and indicate if changes were made. The images or other third partymaterial in this article are included in the article’s Creative Commons license, unlessindicated otherwise in a credit line to the material. If material is not included in thearticle’s Creative Commons license and your intended use is not permitted by statutoryregulation or exceeds the permitted use, you will need to obtain permission directly fromthe copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.

© The Author(s) 2020



https://doi.org/10.1038/s42003-020-0763-1

https://doi.org/10.1038/s42003-020-0763-1

http://www.nature.com/reprints

http://creativecommons.org/licenses/by/4.0/

http://creativecommons.org/licenses/by/4.0/


Recurrent pregnancy loss is associated with a pro...

Documents

Transcript of Recurrent pregnancy loss is associated with a pro...